Imperfect size invariance in face discrimination

Read full text
Authors: Olga Kreichman, Shlomit Zorani, Sharon Gilaie‐Dotan
Cited by: 0

Abstract

Multiple findings suggest that human visual perception exhibit extensive size invariance (i.e., insensitivity to retinal image size changes) for multiple object categories. This is often investigated when retinal image size changes following real world image size changes but viewing distance is kept constant, a situation that does not reflect daily vision where objects retain their real-world size and their image size on the retina changes as a function of viewing distance. Here we measured face discrimination performance to parametrically growing face differences, and assessed size invariance by examining whether performance was sensitive to retinal image size changes employing multiple methods (e.g. human performance deviation from that of the ideal observer). Importantly, retinal image size changes were applied either naturalistically (when faces kept their world size but were more distant) or lab-like (where faces changed their world size while viewing distance was constant). We found that (i) the difference detection threshold (DT) was not influenced by retinal size changes or viewing distances we manipulated, (ii) face discrimination was significantly worse for smaller retinal image sizes (for both naturalistic and for lab-like conditions), (iii) worse face discrimination performance at smaller retinal image sizes were significantly associated between naturalistic and lab-like conditions. Our results indicate that face discrimination shows imperfect invariance to retinal size and importantly, this appears independent of viewing distance. Since face perception is may rely on face-specific mechanisms, future work should examine whether these results generalize to additional object categories.

Fields of Study & Subjects

Face Recognition and Perception Visual perception and processing mechanisms Face recognition and analysis

Publication details

Publication date
6 February 2026
Document type
Preprint